[ThroughMLIR] basic printf support (#1687)

felixdaas · web-flow · commit 1ca3376c89d9 · 2025-06-17T16:54:05.000-07:00
This PR is related to #1685 and adds some basic support for the printf function. Limitations: 1. It only works if all variadic params are of basic interger/float type (for more info why memref type operands don't work see #1685) 2. Only works if the format string is definied directly inside the printf function The downside of this PR is also that the handling this edge case adds significant code bloat and reduces readability for the cir.call op lowering (I tried to insert some meanigful comments to improve the readability), but I think its worth to do this so we have some basic printf support (without adding an extra cir operation) until upstream support for variadic functions is added to the func dialect. Also a few more test (which use such a basic form of printf) in the llvm Single Source test suite are working with this PR: before this PR: Testing Time: 4.00s Total Discovered Tests: 1833 Passed : 420 (22.91%) Failed : 10 (0.55%) Executable Missing: 1403 (76.54%) with this PR: Testing Time: 10.29s Total Discovered Tests: 1833 Passed : 458 (24.99%) Failed : 6 (0.33%) Executable Missing: 1369 (74.69%)
diff --git a/clang/lib/CIR/Lowering/ThroughMLIR/LowerCIRToMLIR.cpp b/clang/lib/CIR/Lowering/ThroughMLIR/LowerCIRToMLIR.cpp
@@ -23,10 +23,12 @@
 #include "mlir/Dialect/ControlFlow/IR/ControlFlowOps.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "mlir/Dialect/LLVMIR/LLVMDialect.h"
+#include "mlir/Dialect/LLVMIR/LLVMTypes.h"
 #include "mlir/Dialect/Math/IR/Math.h"
 #include "mlir/Dialect/MemRef/IR/MemRef.h"
 #include "mlir/Dialect/SCF/IR/SCF.h"
 #include "mlir/Dialect/Vector/IR/VectorOps.h"
+#include "mlir/IR/BuiltinAttributes.h"
 #include "mlir/IR/BuiltinDialect.h"
 #include "mlir/IR/BuiltinOps.h"
 #include "mlir/IR/BuiltinTypes.h"
@@ -104,9 +106,124 @@ class CIRCallOpLowering : public mlir::OpConversionPattern<cir::CallOp> {
     if (mlir::failed(
             getTypeConverter()->convertTypes(op.getResultTypes(), types)))
       return mlir::failure();
-    rewriter.replaceOpWithNewOp<mlir::func::CallOp>(
-        op, op.getCalleeAttr(), types, adaptor.getOperands());
-    return mlir::LogicalResult::success();
+
+    if (!op.isIndirect()) {
+      // Currently variadic functions are not supported by the builtin func
+      // dialect. For now only basic call to printf are supported by using the
+      // llvmir dialect.
+      // TODO: remove this and add support for variadic function calls once
+      // TODO: supported by the func dialect
+      if (op.getCallee()->equals_insensitive("printf")) {
+        SmallVector<mlir::Type> operandTypes =
+            llvm::to_vector(adaptor.getOperands().getTypes());
+
+        // Drop the initial memref operand type (we replace the memref format
+        // string with equivalent llvm.mlir ops)
+        operandTypes.erase(operandTypes.begin());
+
+        // Check that the printf attributes can be used in llvmir dialect (i.e
+        // they have integer/float type)
+        if (!llvm::all_of(operandTypes, [](mlir::Type ty) {
+              return mlir::LLVM::isCompatibleType(ty);
+            })) {
+          return op.emitError()
+                 << "lowering of printf attributes having a type that is "
+                    "converted to memref in cir-to-mlir lowering (e.g. "
+                    "pointers) not supported yet";
+        }
+
+        // Currently only versions of printf are supported where the format
+        // string is defined inside the printf ==> the lowering of the cir ops
+        // will match:
+        // %global = memref.get_global %frm_str
+        // %* = memref.reinterpret_cast (%global, 0)
+        if (auto reinterpret_castOP =
+                mlir::dyn_cast_or_null<mlir::memref::ReinterpretCastOp>(
+                    adaptor.getOperands()[0].getDefiningOp())) {
+          if (auto getGlobalOp =
+                  mlir::dyn_cast_or_null<mlir::memref::GetGlobalOp>(
+                      reinterpret_castOP->getOperand(0).getDefiningOp())) {
+            mlir::ModuleOp parentModule = op->getParentOfType<mlir::ModuleOp>();
+
+            auto context = rewriter.getContext();
+
+            // Find the memref.global op defining the frm_str
+            auto globalOp = parentModule.lookupSymbol<mlir::memref::GlobalOp>(
+                getGlobalOp.getNameAttr());
+
+            rewriter.setInsertionPoint(globalOp);
+
+            // Insert a equivalent llvm.mlir.global
+            auto initialvalueAttr =
+                mlir::dyn_cast_or_null<mlir::DenseIntElementsAttr>(
+                    globalOp.getInitialValueAttr());
+
+            auto type = mlir::LLVM::LLVMArrayType::get(
+                mlir::IntegerType::get(context, 8),
+                initialvalueAttr.getNumElements());
+
+            auto llvmglobalOp = rewriter.create<mlir::LLVM::GlobalOp>(
+                globalOp->getLoc(), type, true, mlir::LLVM::Linkage::Internal,
+                "printf_format_" + globalOp.getSymName().str(),
+                initialvalueAttr, 0);
+
+            rewriter.setInsertionPoint(getGlobalOp);
+
+            // Insert llvmir dialect ops to retrive the !llvm.ptr of the global
+            auto globalPtrOp = rewriter.create<mlir::LLVM::AddressOfOp>(
+                getGlobalOp->getLoc(), llvmglobalOp);
+
+            mlir::Value cst0 = rewriter.create<mlir::LLVM::ConstantOp>(
+                getGlobalOp->getLoc(), rewriter.getI8Type(),
+                rewriter.getIndexAttr(0));
+            auto gepPtrOp = rewriter.create<mlir::LLVM::GEPOp>(
+                getGlobalOp->getLoc(),
+                mlir::LLVM::LLVMPointerType::get(context),
+                llvmglobalOp.getType(), globalPtrOp,
+                ArrayRef<mlir::Value>({cst0, cst0}));
+
+            mlir::ValueRange operands = adaptor.getOperands();
+
+            // Replace the old memref operand with the !llvm.ptr for the frm_str
+            mlir::SmallVector<mlir::Value> newOperands;
+            newOperands.push_back(gepPtrOp);
+            newOperands.append(operands.begin() + 1, operands.end());
+
+            // Create the llvmir dialect function type for printf
+            auto llvmI32Ty = mlir::IntegerType::get(context, 32);
+            auto llvmPtrTy = mlir::LLVM::LLVMPointerType::get(context);
+            auto llvmFnType =
+                mlir::LLVM::LLVMFunctionType::get(llvmI32Ty, llvmPtrTy,
+                                                  /*isVarArg=*/true);
+
+            rewriter.setInsertionPoint(op);
+
+            // Insert an llvm.call op with the updated operands to printf
+            rewriter.replaceOpWithNewOp<mlir::LLVM::CallOp>(
+                op, llvmFnType, op.getCalleeAttr(), newOperands);
+
+            // Cleanup printf frm_str memref ops
+            rewriter.eraseOp(reinterpret_castOP);
+            rewriter.eraseOp(getGlobalOp);
+            rewriter.eraseOp(globalOp);
+
+            return mlir::LogicalResult::success();
+          }
+        }
+
+        return op.emitError()
+               << "lowering of printf function with Format-String"
+                  "defined outside of printf is not supported yet";
+      }
+
+      rewriter.replaceOpWithNewOp<mlir::func::CallOp>(
+          op, op.getCalleeAttr(), types, adaptor.getOperands());
+      return mlir::LogicalResult::success();
+
+    } else {
+      // TODO: support lowering of indirect calls via func.call_indirect op
+      return op.emitError() << "lowering of indirect calls not supported yet";
+    }
   }
 };
 
@@ -557,37 +674,60 @@ class CIRFuncOpLowering : public mlir::OpConversionPattern<cir::FuncOp> {
                   mlir::ConversionPatternRewriter &rewriter) const override {
 
     auto fnType = op.getFunctionType();
-    mlir::TypeConverter::SignatureConversion signatureConversion(
-        fnType.getNumInputs());
 
-    for (const auto &argType : enumerate(fnType.getInputs())) {
-      auto convertedType = typeConverter->convertType(argType.value());
-      if (!convertedType)
-        return mlir::failure();
-      signatureConversion.addInputs(argType.index(), convertedType);
-    }
+    if (fnType.isVarArg()) {
+      // TODO: once the func dialect supports variadic functions rewrite this
+      // For now only insert special handling of printf via the llvmir dialect
+      if (op.getSymName().equals_insensitive("printf")) {
+        auto context = rewriter.getContext();
+        // Create a llvmir dialect function declaration for printf, the
+        // signature is: i32 (!llvm.ptr, ...)
+        auto llvmI32Ty = mlir::IntegerType::get(context, 32);
+        auto llvmPtrTy = mlir::LLVM::LLVMPointerType::get(context);
+        auto llvmFnType =
+            mlir::LLVM::LLVMFunctionType::get(llvmI32Ty, llvmPtrTy,
+                                              /*isVarArg=*/true);
+        auto printfFunc = rewriter.create<mlir::LLVM::LLVMFuncOp>(
+            op.getLoc(), "printf", llvmFnType);
+        rewriter.replaceOp(op, printfFunc);
+      } else {
+        rewriter.eraseOp(op);
+        return op.emitError() << "lowering of variadic functions (except "
+                                 "printf) not supported yet";
+      }
+    } else {
+      mlir::TypeConverter::SignatureConversion signatureConversion(
+          fnType.getNumInputs());
+
+      for (const auto &argType : enumerate(fnType.getInputs())) {
+        auto convertedType = typeConverter->convertType(argType.value());
+        if (!convertedType)
+          return mlir::failure();
+        signatureConversion.addInputs(argType.index(), convertedType);
+      }
 
-    SmallVector<mlir::NamedAttribute, 2> passThroughAttrs;
+      SmallVector<mlir::NamedAttribute, 2> passThroughAttrs;
 
-    if (auto symVisibilityAttr = op.getSymVisibilityAttr())
-      passThroughAttrs.push_back(
-          rewriter.getNamedAttr("sym_visibility", symVisibilityAttr));
+      if (auto symVisibilityAttr = op.getSymVisibilityAttr())
+        passThroughAttrs.push_back(
+            rewriter.getNamedAttr("sym_visibility", symVisibilityAttr));
 
-    mlir::Type resultType =
-        getTypeConverter()->convertType(fnType.getReturnType());
-    auto fn = rewriter.create<mlir::func::FuncOp>(
-        op.getLoc(), op.getName(),
-        rewriter.getFunctionType(signatureConversion.getConvertedTypes(),
-                                 resultType ? mlir::TypeRange(resultType)
-                                            : mlir::TypeRange()),
-        passThroughAttrs);
+      mlir::Type resultType =
+          getTypeConverter()->convertType(fnType.getReturnType());
+      auto fn = rewriter.create<mlir::func::FuncOp>(
+          op.getLoc(), op.getName(),
+          rewriter.getFunctionType(signatureConversion.getConvertedTypes(),
+                                   resultType ? mlir::TypeRange(resultType)
+                                              : mlir::TypeRange()),
+          passThroughAttrs);
 
-    if (failed(rewriter.convertRegionTypes(&op.getBody(), *typeConverter,
-                                           &signatureConversion)))
-      return mlir::failure();
-    rewriter.inlineRegionBefore(op.getBody(), fn.getBody(), fn.end());
+      if (failed(rewriter.convertRegionTypes(&op.getBody(), *typeConverter,
+                                             &signatureConversion)))
+        return mlir::failure();
+      rewriter.inlineRegionBefore(op.getBody(), fn.getBody(), fn.end());
 
-    rewriter.eraseOp(op);
+      rewriter.eraseOp(op);
+    }
     return mlir::LogicalResult::success();
   }
 };
diff --git a/clang/test/CIR/Lowering/ThroughMLIR/call.c b/clang/test/CIR/Lowering/ThroughMLIR/call.c
@@ -12,3 +12,41 @@ int test(void) {
 //       CHECK:   %[[ARG:.+]] = arith.constant 2 : i32
 //  CHECK-NEXT:   call @foo(%[[ARG]]) : (i32) -> ()
 //       CHECK: }
+
+extern int printf(const char *str, ...);
+
+// CHECK-LABEL: llvm.func @printf(!llvm.ptr, ...) -> i32
+//       CHECK: llvm.mlir.global internal constant @[[FRMT_STR:.*]](dense<[37, 100, 44, 32, 37, 102, 44, 32, 37, 100, 44, 32, 37, 108, 108, 100, 44, 32, 37, 100, 44, 32, 37, 102, 10, 0]> : tensor<26xi8>) {addr_space = 0 : i32} : !llvm.array<26 x i8>
+
+void testfunc(short s, float X, char C, long long LL, int I, double D) {
+	printf("%d, %f, %d, %lld, %d, %f\n", s, X, C, LL, I, D);
+}
+
+// CHECK: func.func @testfunc(%[[ARG0:.*]]: i16 {{.*}}, %[[ARG1:.*]]: f32 {{.*}}, %[[ARG2:.*]]: i8 {{.*}}, %[[ARG3:.*]]: i64 {{.*}}, %[[ARG4:.*]]: i32 {{.*}}, %[[ARG5:.*]]: f64 {{.*}}) {
+// CHECK: %[[ALLOCA_S:.*]] = memref.alloca() {alignment = 2 : i64} : memref<i16>
+// CHECK: %[[ALLOCA_X:.*]] = memref.alloca() {alignment = 4 : i64} : memref<f32>  
+// CHECK: %[[ALLOCA_C:.*]] = memref.alloca() {alignment = 1 : i64} : memref<i8> 
+// CHECK: %[[ALLOCA_LL:.*]] = memref.alloca() {alignment = 8 : i64} : memref<i64> 
+// CHECK: %[[ALLOCA_I:.*]] = memref.alloca() {alignment = 4 : i64} : memref<i32> 
+// CHECK: %[[ALLOCA_D:.*]] = memref.alloca() {alignment = 8 : i64} : memref<f64> 
+// CHECK: memref.store %[[ARG0]], %[[ALLOCA_S]][] : memref<i16> 
+// CHECK: memref.store %[[ARG1]], %[[ALLOCA_X]][] : memref<f32>
+// CHECK: memref.store %[[ARG2]], %[[ALLOCA_C]][] : memref<i8> 
+// CHECK: memref.store %[[ARG3]], %[[ALLOCA_LL]][] : memref<i64> 
+// CHECK: memref.store %[[ARG4]], %[[ALLOCA_I]][] : memref<i32> 
+// CHECK: memref.store %[[ARG5]], %[[ALLOCA_D]][] : memref<f64> 
+// CHECK: %[[FRMT_STR_ADDR:.*]] = llvm.mlir.addressof @[[FRMT_STR]] : !llvm.ptr 
+// CHECK: %[[C0:.*]] = llvm.mlir.constant(0 : index) : i8 
+// CHECK: %[[FRMT_STR_DATA:.*]] = llvm.getelementptr %[[FRMT_STR_ADDR]][%[[C0]], %[[C0]]] : (!llvm.ptr, i8, i8) -> !llvm.ptr, !llvm.array<26 x i8> 
+// CHECK: %[[S:.*]] = memref.load %[[ALLOCA_S]][] : memref<i16> 
+// CHECK: %[[S_EXT:.*]] = arith.extsi %3 : i16 to i32 
+// CHECK: %[[X:.*]] = memref.load %[[ALLOCA_X]][] : memref<f32> 
+// CHECK: %[[X_EXT:.*]] = arith.extf %5 : f32 to f64 
+// CHECK: %[[C:.*]] = memref.load %[[ALLOCA_C]][] : memref<i8> 
+// CHECK: %[[C_EXT:.*]] = arith.extsi %7 : i8 to i32
+// CHECK: %[[LL:.*]] = memref.load %[[ALLOCA_LL]][] : memref<i64> 
+// CHECK: %[[I:.*]] = memref.load %[[ALLOCA_I]][] : memref<i32> 
+// CHECK: %[[D:.*]] = memref.load %[[ALLOCA_D]][] : memref<f64> 
+// CHECK: {{.*}} = llvm.call @printf(%[[FRMT_STR_DATA]], %[[S_EXT]], %[[X_EXT]], %[[C_EXT]], %[[LL]], %[[I]], %[[D]]) vararg(!llvm.func<i32 (ptr, ...)>) : (!llvm.ptr, i32, f64, i32, i64, i32, f64) -> i32 
+// CHECK: return
+// CHECK: } 
